Probabilistic Approaches to Population Forecasting

ON AVERAGE people live long lives, have a long lag between birth and childbearing, and experience demographic rates with highly regular age patterns. These patterns generally change quite slowly. For these reasons, population is reliably predictable over fairly long periods compared to economic performance or the weather. Nonetheless, demographic forecasting does involve a good deal of uncertainty. We know that the US population will age rapidly between 2010 and 2035 as the huge baby boom generations move into old age; but we do not know what share of these generations will survive to old age, or how many potential workers will be born and survive throughout their working years to help support the elderly. We all know that demographic forecasts are often seriously wrong, and realize that forecasting errors are inevitable. It is generally agreed that demographers have a responsibility to indicate how certain or uncertain their forecasts may be. Traditional forecasts assess and communicate the uncertainty surrounding the middle, or preferred, forecast through the use of high and low versions. Each scenario, or set of assumptions underlying one of these three versions, contains an assumed trajectory for fertility, another for mortality, and another for migration. These scenario-based indications of uncertainty are of some use, but they have certain serious problems: no probability is attached to their high-low ranges, and they are internally inconsistent in the sense that they misrepresent the relative uncertainty in different measures such as population size, fertility, and old-age dependency ratios, for reasons that will be explained later. One visible and inevitable sign of this problem is that when the high-low scenarios are chosen to bracket the long-term population growth or age distribution, the annual values of fertility or births often fall outside the high-low range soon after the forecasts are published, making the forecasters appear (unjustly) to be incompetent. Probabilistic population forecasts offer an alternative approach to assessing and communicating uncertainty. "Probabilistic forecasts" can be un-